Prediction of human cardiotoxic QT prolongation using in-vitro multiple ion channel data and mathematical models of cardiac myocytes
Lead Research Organisation:
University of Oxford
Department Name: Computer Science
Abstract
The Problem
The leading cause of withdrawal of pharmaceutical drugs from the market is a disturbance to the heart's rhythm, which occurs rarely, but can cause sudden cardiac death. This is very difficult to predict, so to detect the risk of this happening, experiments are performed by pharmaceutical companies during the development of new drugs. Currently, thousands of animal tests are needed to detect the risk of these new drugs causing side effects on the heart. There are many different experiments, including tests on tissues removed from animals, and later on, studies on conscious animals. A major problem is that these animal experiments give accurate predictions of what will happen in humans in only around 70% of cases.
Background
The reason drugs can cause side effects on the heart is that the heart's activity depends on the flow of electrically charged particles - known as ions - in and out of heart muscle cells through channels made of proteins. These ion channels can be blocked by pharmaceutical drugs, and this can cause unwanted disturbances to the heart's rhythm, called arrhythmias. Recent innovations in cell line technologies mean that pharmaceutical companies can now detect to what extent a drug blocks an ion channel as a matter of routine, without using animal experiments. By repeating these experiments for many different ion channels we can screen all of the susceptible ion channels in the heart, and measure the amount of ion channel blocking that occurs as we increase the drug dose.
Our Project
In this project, we will perform these ion channel screens and use the results to feed into mathematical models for the electrical activity of the heart. We will simulate the effect of a drug on a single heart cell, and then the whole heart, predicting whether a set of drugs will cause harmful side effects. To evaluate our success we will compare our predictions with the results of human clinical trials.
Benefits
These computer simulations are very fast and cheap in comparison with animal experiments, enabling a risk assessment to be performed for more drugs, earlier in development. Since animal experiments are not perfect predictors of the results of human trials we may also be able to improve on their results by using mathematical models of human cells. In doing so, we aim to reduce and replace the current use of animal experiments in assessing drug-induced risks to the heart with computer simulations, and also to make more accurate predictions of what will happen when the drug is given to people.
The leading cause of withdrawal of pharmaceutical drugs from the market is a disturbance to the heart's rhythm, which occurs rarely, but can cause sudden cardiac death. This is very difficult to predict, so to detect the risk of this happening, experiments are performed by pharmaceutical companies during the development of new drugs. Currently, thousands of animal tests are needed to detect the risk of these new drugs causing side effects on the heart. There are many different experiments, including tests on tissues removed from animals, and later on, studies on conscious animals. A major problem is that these animal experiments give accurate predictions of what will happen in humans in only around 70% of cases.
Background
The reason drugs can cause side effects on the heart is that the heart's activity depends on the flow of electrically charged particles - known as ions - in and out of heart muscle cells through channels made of proteins. These ion channels can be blocked by pharmaceutical drugs, and this can cause unwanted disturbances to the heart's rhythm, called arrhythmias. Recent innovations in cell line technologies mean that pharmaceutical companies can now detect to what extent a drug blocks an ion channel as a matter of routine, without using animal experiments. By repeating these experiments for many different ion channels we can screen all of the susceptible ion channels in the heart, and measure the amount of ion channel blocking that occurs as we increase the drug dose.
Our Project
In this project, we will perform these ion channel screens and use the results to feed into mathematical models for the electrical activity of the heart. We will simulate the effect of a drug on a single heart cell, and then the whole heart, predicting whether a set of drugs will cause harmful side effects. To evaluate our success we will compare our predictions with the results of human clinical trials.
Benefits
These computer simulations are very fast and cheap in comparison with animal experiments, enabling a risk assessment to be performed for more drugs, earlier in development. Since animal experiments are not perfect predictors of the results of human trials we may also be able to improve on their results by using mathematical models of human cells. In doing so, we aim to reduce and replace the current use of animal experiments in assessing drug-induced risks to the heart with computer simulations, and also to make more accurate predictions of what will happen when the drug is given to people.
Technical Summary
Thousands of animals are used across the world for the assessment of cardiac toxicity each year. Animals are used at multiple stages of drug development, in every pharmaceutical company (Pollard et al., 2010). This is primarily for detection of risk of Torsade-de-Pointes (TdP) cardiac arrhythmia. A leading cause of withdrawal of drugs from the market, TdP risk is one of the main causes of attrition during compound development. There are two major reasons that large numbers of animals have traditionally been required. (1) There are a large number of potential drug interactions in the heart, which we could not hope to screen without a representation of all of the possible targets in the whole system (with an animal model); and (2) the heart's electrophysiology has been considered "too complicated" to predict a drug effect - even given the full list of drug targets and affinities, the whole physiological system must be well represented (again, with an animal model).
Technological advances mean that neither of the points above should remain a stumbling block, and in this project we will reduce animal use by taking advantage of the following techniques: (1) We will work with AstraZeneca (AZ) and GlaxoSmithKline (GSK) to assess compounds for multiple cardiac-ion channel interactions, using high-throughput in-vitro screens, to address the first point; (2) mathematical models, quantifying the complex processes involved in generation of cardiac electrical activity, address the second. We will compare our predictions with the human trial results, statistically quantifying the level of predictive power that simulations have for human clinical trials. We will provide all of the generated data, simulation and analysis tools as open-source. There will therefore be no major obstacle to the widespread use of simulation, instead of animal models, for pro-arrhythmic screening, with additional benefits in terms of more accurate prediction of effects in human physiology.
Technological advances mean that neither of the points above should remain a stumbling block, and in this project we will reduce animal use by taking advantage of the following techniques: (1) We will work with AstraZeneca (AZ) and GlaxoSmithKline (GSK) to assess compounds for multiple cardiac-ion channel interactions, using high-throughput in-vitro screens, to address the first point; (2) mathematical models, quantifying the complex processes involved in generation of cardiac electrical activity, address the second. We will compare our predictions with the human trial results, statistically quantifying the level of predictive power that simulations have for human clinical trials. We will provide all of the generated data, simulation and analysis tools as open-source. There will therefore be no major obstacle to the widespread use of simulation, instead of animal models, for pro-arrhythmic screening, with additional benefits in terms of more accurate prediction of effects in human physiology.
Planned Impact
Pro-arrhythmic safety testing consists of many different animal-based experiments, at different stages of drug development. These increase in cost and complexity, and correspondingly decrease in throughput, as potential drug candidates are selected. A typical set of experiments for assessment of QT prolongation includes in-vitro isolated cell and ex-vivo tissue preparations, together with in-vivo conscious rodent QT telemetry and conscious higher mammal QT telemetry. In AZ alone around 1300 guinea pigs and dogs were used for QT assessment during 2009-2011, translating to over 10,000 animals a year across the sector, with a significant proportion used in in-vivo studies.
There are two main ways in which the 3Rs impact of this project will be maximised. 1) We will perform screens, simulations and a statistical evaluation, with an unprecedented number of compounds, to provide confidence in replacement via simulation. 2) We will provide simulation tools, deploying them inside two companies, and providing them open-source to others. We expand upon these points below.
1) We aim to thoroughly evaluate the degree to which computational predictions replicate human QT prolongation clinical trial results. This has never been done before; previous studies were limited to a handful of compounds at most. Our unique position will take advantage of the new technologies and available expertise at AZ and GSK. The overall aim is to reduce the number of conscious in-vivo animal experiments that occur. As human QT prolongation is being more accurately predicted, earlier in drug development (see Figure 2 in Case for Support), the number of animals used can be reduced, and the emphasis of remaining experiments can be shifted from detection of QT to improved detection of other side-effects.
Note that in addition to simulation being used to replace in-vivo animal experiments, it can also be used to reduce the number of animal experiments: detection of a QT positive compound at the early stages of drug development means a compound can be confidently discarded, removing it from animal testing stages altogether. The remaining animal tests that are conducted are more likely to be free of detrimental side effects, and to lead to a marketable drug.
The replacement of in-vitro tests, whilst not in the direct scope of the 3Rs, is another outcome of this work. Should our predictions provide sufficient sensitivity and specificity for human trial results, then certain assays will be replaced immediately. In particular, we are examining whether the GSK rabbit left ventricular wedge QT study, which uses over 400 animals per year, could be replaced by simulation. The AZ isolated myocyte study has already been scaled down (from 23 to 1 dog per year) via the simulation approach (Davies et al., 2012). Use of simulation to reproduce the results of these in-vitro/ex-vivo assays provides confidence that simulations we will develop during the proposed project will be capable of reducing and replacing in-vivo animal tests.
2) As part of this project we will be developing and providing open-source software for a simulation back-end, and a web-portal front-end. A prototype of these is currently under development, and is presently being trialled inside GSK. Although the accuracy of simulations has not yet been established (this project will provide the first large body of evidence) the safety pharmacologists at GSK find the role of simulation in combining quantitative screening information to be highly useful. This experience provides us with confidence that a robust software package, compliant with security standards, can be developed and deployed within the timescale of this project. It has also become apparent that such considerations are crucial to the uptake of the technology. It is for this reason that almost half of the requested resources are directly allocated to implementation and deployment, alongside the scientific work.
There are two main ways in which the 3Rs impact of this project will be maximised. 1) We will perform screens, simulations and a statistical evaluation, with an unprecedented number of compounds, to provide confidence in replacement via simulation. 2) We will provide simulation tools, deploying them inside two companies, and providing them open-source to others. We expand upon these points below.
1) We aim to thoroughly evaluate the degree to which computational predictions replicate human QT prolongation clinical trial results. This has never been done before; previous studies were limited to a handful of compounds at most. Our unique position will take advantage of the new technologies and available expertise at AZ and GSK. The overall aim is to reduce the number of conscious in-vivo animal experiments that occur. As human QT prolongation is being more accurately predicted, earlier in drug development (see Figure 2 in Case for Support), the number of animals used can be reduced, and the emphasis of remaining experiments can be shifted from detection of QT to improved detection of other side-effects.
Note that in addition to simulation being used to replace in-vivo animal experiments, it can also be used to reduce the number of animal experiments: detection of a QT positive compound at the early stages of drug development means a compound can be confidently discarded, removing it from animal testing stages altogether. The remaining animal tests that are conducted are more likely to be free of detrimental side effects, and to lead to a marketable drug.
The replacement of in-vitro tests, whilst not in the direct scope of the 3Rs, is another outcome of this work. Should our predictions provide sufficient sensitivity and specificity for human trial results, then certain assays will be replaced immediately. In particular, we are examining whether the GSK rabbit left ventricular wedge QT study, which uses over 400 animals per year, could be replaced by simulation. The AZ isolated myocyte study has already been scaled down (from 23 to 1 dog per year) via the simulation approach (Davies et al., 2012). Use of simulation to reproduce the results of these in-vitro/ex-vivo assays provides confidence that simulations we will develop during the proposed project will be capable of reducing and replacing in-vivo animal tests.
2) As part of this project we will be developing and providing open-source software for a simulation back-end, and a web-portal front-end. A prototype of these is currently under development, and is presently being trialled inside GSK. Although the accuracy of simulations has not yet been established (this project will provide the first large body of evidence) the safety pharmacologists at GSK find the role of simulation in combining quantitative screening information to be highly useful. This experience provides us with confidence that a robust software package, compliant with security standards, can be developed and deployed within the timescale of this project. It has also become apparent that such considerations are crucial to the uptake of the technology. It is for this reason that almost half of the requested resources are directly allocated to implementation and deployment, alongside the scientific work.
